CN108983051A - Based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation - Google Patents
Based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation Download PDFInfo
- Publication number
- CN108983051A CN108983051A CN201810829627.6A CN201810829627A CN108983051A CN 108983051 A CN108983051 A CN 108983051A CN 201810829627 A CN201810829627 A CN 201810829627A CN 108983051 A CN108983051 A CN 108983051A
- Authority
- CN
- China
- Prior art keywords
- signal
- imt
- synchronous
- frequency
- wavelet
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Relating To Insulation (AREA)
Abstract
It based on the purpose of synchronous shelf depreciation kind identification method for squeezing wavelet transformation is a kind of recognition methods proposed for transformer partial discharge signal that the present invention, which provides a kind of, realize the identification of transformer partial discharge signal, firstly, being decomposed using the synchronous wavelet transformation that squeezes to typical transformer partial discharge signal;Then, using the difference of local discharge signal energy and complexity on different decomposition scale, the characteristic quantity identified using multiple dimensioned arrangement entropy as electric discharge type;Finally, the characteristic quantity support vector machine classifier extracted is carried out discharge mode identification.The local discharge signal of this patent method is averaged recognition accuracy higher than 90%, hence it is evident that is better than other common partial discharge of transformer recognition methods.
Description
Technical field
The present invention relates to a kind of based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation, power system signal
Process field.
Background technique
Power transformer is one of the equipment of most critical in electric system, and the quality of security performance affects the peace of power grid
Complete effectively operation.Shelf depreciation (Partial Discharge, PD) is to cause the apparatus insulated damage of the Large-scale High-Pressures such as power transformer
Bad the main reason for, and different shelf depreciation types are different to the degree of insulation damage, the mechanism formed is also different.Cause
This, rapidly and accurately identifies different partial discharge of transformer types, both provides for the subsequent differentiation for carrying out abort situation
Solid strong foundation, also effectively running to the stabilization of maintenance electric system has important directive significance.
Currently, the method for detection device shelf depreciation specifically includes that pulse current method, hyperfrequency method, supercritical ultrasonics technology, optics
Method, chemical method etc..Wherein, since hyperfrequency method (Ultra-High Frequency, UHF) has in-site installation convenient, sensitive
It the advantages that degree height, strong antijamming capability, is widely applied locally putting in single on-line monitoring.
Due to the complexity of field monitoring environment, single high-frequency local discharging (UHF PD) signal obtained from detection itself
It sets out, it tends to be difficult to the accurate differentiation for realizing defect type.Therefore, it is necessary to carry out feature to the UHF PD signal detected to mention
It takes, acquisition can effectively distinguish the characteristic quantity of all kinds of defects, to realize the identification of defect type.Effective feature extracting method
It is the basis for carrying out defect recognition, the selection of characteristic quantity directly affects the accuracy of recognition result.
Currently, the feature extracting method of PD signal is broadly divided into two major classes, one kind is statistics spectrogram method, multiple by acquiring
The PD signal of power frequency period, construction two dimension or three-dimensional statistics spectrogram, then therefrom extract statistical nature, fractal characteristic, digital picture
The characteristic parameters such as feature.For UHF PD signal, according to statistics spectrogram method, sample rate requires height, and data volume is big, processing
Data speed is slow, is unfavorable for monitoring on-line, and constructs statistics spectrogram and need the phase information of PD, but monitoring field often
It is difficult to obtain;Another kind of is Waveform Analysis Method, by acquiring single UHF PD signal waveform, extract the time domain of signal, frequency domain or
Other transform domain features.This method data volume is small, and processing speed is fast, and does not need discharge phase information, but due to PD pulse
The electromagnetic wave of excitation there is decaying in communication process and catadioptric, monitoring field exist simultaneously serious electromagnetic interference, passes
For the characteristic parameter based on time domain or frequency domain of system vulnerable to noise pollution, the characteristic parameter that accurately extract PD signal is relatively difficult.
Wavelet transformation has good Time-Frequency Localization analysis ability, can obtain signal part simultaneously using wavelet transformation
Time domain and frequency domain information, obtain be capable of it is more accurate and it is effective describe signal Analysis On Multi-scale Features parameter, in PD signal characteristic
It is widely applied in extraction.However, causing radio-frequency head to divide since wavelet transformation only does further decomposition to low frequency part
Rate resolution ratio is poor.There is frequency range of interlocking in PD signal amplitude-frequency response after wavelet transformation decomposes, intersubband often exists seriously
Spectral aliasing and energy leakage, thus the Analysis On Multi-scale Features parameter extracted from the subband of wavelet decomposition tends not to accurately retouch
The Time-Frequency Information for stating PD signal is unfavorable for subsequent Classification and Identification.
Summary of the invention
The purpose of the present invention is to provide a kind of based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation, with
It solves the above problems.
Present invention employs following technical solutions:
A kind of shelf depreciation kind identification method based on synchronous extruding wavelet transformation characterized by comprising
Step 1: being decomposed using the synchronous wavelet transformation that squeezes to 4 kinds of typical transformer partial discharge signals, utilize
SWT carries out accurate careful division to time frequency plane, and decomposition obtains subband signal
Generally can be analyzed to the superposition of multiple eigenfunctions, i.e. signal f (t) is represented by time varying signal f (t)
In formula: AkIt (t) is the instantaneous amplitude of k-th of component;AkIt (t) is the instantaneous amplitude of k-th of component, φkIt (t) is kth
The instantaneous phase of a component;R (t) is noise or error, and K indicates the component number of signal, and synchronous extruding wavelet transformation passes through thin
The time-frequency curve for changing wavelet transformation, extracts the amplitude factor A of each componentk(t) and instantaneous frequency φ 'k(t) (k=1,2 ...,
K), the synchronous wavelet transformation SWT that squeezes is according to the coefficient W after wavelet transformationf(a, b) is in (a, b) neighbouring local property by coefficient Wf
(a, b) is reassigned to the difference (ω in time frequency planef(a,b),b)(ωf(a, b) indicates instantaneous frequency of the signal at (a, b)
Rate), to keep time-frequency curve thinner apparent, improves frequency resolution and reduce modal overlap, when so that component signal reconstructing
Precision it is higher,
The synchronous wavelet transformation that squeezes gives wavelet mother function ψ (t), signal f based on the continuous wavelet transform of signal
(t) continuous wavelet transform is
In formula: ψ * indicates the conjugation of mother wavelet function, and a is scale factor, and b is shift factor, fixed according to Plancherel
Reason, formula (2) are in the equivalence transformation of frequency domain
In formula, ξ is pi,It is the Fourier transformation of f (t), ψ (t) respectively, to simplest signal f
(t)=A cos (ω t), Fourier transformation areAccording to formula (3), continuous wavelet
It is transformed to
It is assumed that wavelet function ψ has rapid decay, andIn ξ=ω0Locate integrated distribution, then the wavelet coefficient of signal
Wf(a, b) will be in scalePlace concentrates, but can in a certain range along size distribution,
For any one in wavelet transform result " time-scale " point, (b a) can be estimated by wavelet coefficient derivation
The instantaneous frequency of signal, i.e.,
The synchronous wavelet transformation that squeezes establishes (a, b) → [ω on the basis of instantaneous frequencyf(a, b), b] mapping, will be small
Wave system number Wf(a, b) is transformed into " time-frequency " plane W by " time-scale " planef[ωf(a, b), b], it, will in SWT
Any centre frequency ωlNeighbouring sectionWavelet coefficient values be expressed to centre frequency ωlOn, it obtains
It is synchronous to squeeze transformed value Tf(ωl, b), in calculating, due to a, b, ω is discrete, it is assumed that ai-ai-1=(Δ a)i, it is small to synchronize extruding
Wave conversion value Tf(ωl, b) and it is represented by
Wavelet transformation SWT and wavelet transformation WT, in short-term Fourier's change are squeezed by the way that a time varying signal f (t) is relatively more synchronous
Change the difference of STFT, f (t) is formed by the Signal averaging of three different frequencies: 0-0.7s is the cosine signal f of 20Hz1(t)=
Cos (40 π t), 0.3-1s are the cosine signal f of 30Hz2(t)=cos (60 π t), 0-1s are the cosine that frequency is vibrated in 80Hz
FM signal f3(t)=cos [160 π t-5cos (30t)], while the white Gaussian noise that signal-to-noise ratio is 3dB is added to signal, it is right
Signal after noise is added is utilized respectively empirical mode decomposition, and wavelet package transforms carry out frequency spectrum with synchronous extruding wavelet transformation SWT
Analysis,
The synchronous wavelet transformation that squeezes is reversible, and for multicomponent data processing, passes through Tf(ωl, b) can not only reconstruct it is original
Signal f (t), and can be with each component signal of Accurate Reconstruction fk(t), it is assumed that LkIt (t) is in time-frequency figure with fk(t) crestal line
Centered on a minizone, then fk(t) reconstruction formula is
In formula
Step 2: synchronous squeeze the multiple dimensioned arrangement entropy measure of wavelet transformation SWT
It is synchronous to squeeze small echo using the multiple dimensioned energy-distributing feature of time-frequency domain as the characteristic quantity for distinguishing different type defect
The local characteristics of signal difference vibration frequency are extracted in transformation, calculate it is synchronous squeeze wavelet transformation SWT arrangement moisture in the soil value just it can be found that
Small and very brief exception is defined as follows on the basis of local discharge signal multi-scale Representation along size distribution in signal
The synchronous small echo that squeezes arranges entropy measure,
If accumulateing mode class function in after the synchronized extruding window Fourier transform of local discharge signal is IMTk(k=1,
2 ..., K), to mode class function IMTkPhase space reconfiguration is carried out, can be obtained
In formula: m, τ respectively indicate Embedded dimensions and delay time;The IMT that Q=N- (m-1) τ, N are indicatedkLength, by the square
Battle array every a line as a reconstruct component, then Q reconstruct component can be obtained, in (1) formula j-th of component [IMT (j),
IMT] (j+ τ) ..., IMT (j+ (m-1) τ)], its element is rearranged by increasing mode, if i1,i2,…,imExpression is arranged again
The index of each element position after column has IMT [j+ (i1-1)τ]≤IMT[j+(i2-1)τ]≤…≤IMT[j+(im-1)
τ],
If the value there are two component is equal, i.e.,
IMT[j+(i1- 1) τ]=IMT [j+ (i2-1)τ]
Then in arrangement according to index value i1And i2Size arrange, that is, work as i1< i2When, IMT [j+ (i1- 1) τ] it comes
IMT[j+(i2- 1) τ] before, put in order at this time for
IMT[j+(i1-1)τ]≤IMT[j+(i2-1)τ]
Therefore, for IMTkObtained restructuring matrix after rearranging to each of these row, can obtain
One group based on the symbol sebolic addressing to put in order
S (r)=(i1,i2,…,im), wherein r=1,2 ..., q, and q≤m!
When each row element in restructuring matrix is arranged, if not considering the size arbitrary arrangement of element value,
Its m element shares m!Middle aligning method, i.e., shared m!A symbol sebolic addressing (i1,i2,…,im), it is arranged by element value size
Symbol sebolic addressing S (r) afterwards is one of which, summarizes time that each symbol sebolic addressing S (r) occurs in restructuring matrix arrangement
Number, and calculate its corresponding probability, it is assumed that it is respectively P1,P2,…,Pq, then can according to the definition of Shannon entropy calculate in accumulate
Mode class function IMTkQ kind distinct symbols sequence arrangement entropy Ep(m), i.e.,
When the arrangement S (r) of restructuring matrix disperse the most namely eachWhen, arrangement entropy reaches maximum value Ep(m)
=ln (m!), therefore, arrange entropy Ep(m) size accumulates mode class function IMT in can describingkThe random degree of middle sequence: arrangement
Entropy Ep(m) value is smaller, shows IMTkIn data it is more regular;Arrange entropy Ep(m) value is bigger, then shows IMTkIn data
More irregularly, closer to random sequence, entropy E is arrangedp(m) variation, which can reflect and be exaggerated in multiple dimensioned, accumulates mode class letter
Number IMTkThe variations in detail of middle data sequence can go out the mutation of sequence with accurate detection, can get in shelf depreciation classification
Higher resolution ratio, for the ease of using, it usually needs to Ep(m) it is normalized, evenThen 0≤Ep
≤ 1,
Step 3: carrying out the knowledge of local discharge signal defect type using support vector machines according to the arrangement entropy of extraction
Not.
Further, of the invention based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation, also have in this way
Feature:
In step 3, the most common gaussian radial basis function of the Selection of kernel function of SVM, expression formula is
In identification process, SVM uses one-to-one more disaggregated models, and determines SVM most using 2- folding cross validation method
Good regularization coefficient C=0.3 and kernel functional parameter σ=0.65.
Advantageous effect of the invention
Shelf depreciation kind identification method based on synchronous extruding wavelet transformation of the invention, passes through synchronous extruding transformation pair
The decomposition of local discharge signal, the shortcomings that overcoming wavelet decomposition intersubband there are spectral aliasing and energy leakages, what is used is more
Scale arrangement entropy characteristic parameter can effectively portray UHF PD signal in the Energy distribution and complexity information of time-frequency domain, have compared with
Good stability and anti-interference ability.
The local discharge signal of this patent method is averaged recognition accuracy higher than 90%, hence it is evident that is better than other common transformations
Device Recognition of Partial Discharge.
Detailed description of the invention
Fig. 1 is the time frequency analysis result figure for adding cosine frequency modulation signals of making an uproar;
Fig. 2 (a) is suspended discharge model;
Fig. 2 (b) is needle plate discharging model;
Fig. 2 (c) is creeping discharge model;
Fig. 2 (d) is bubble-discharge model;
Fig. 3 is the construction process in UHF PD signal characteristic library;
Fig. 4 is 95% confidence interval of the multiple dimensioned arrangement entropy of 4 kinds of UHF PD signals.
Specific embodiment
Illustrate a specific embodiment of the invention below in conjunction with attached drawing.
Step 1: being decomposed using the synchronous wavelet transformation that squeezes to 4 kinds of typical transformer partial discharge signals, utilize
The synchronous wavelet transformation SWT that squeezes can carry out accurate careful division, the subband signal decomposed to time frequency plane
Generally can be analyzed to the superposition of multiple eigenfunctions, i.e. signal f (t) is represented by time varying signal f (t)
In formula: AkIt (t) is the instantaneous amplitude of k-th of component;AkIt (t) is the instantaneous amplitude of k-th of component, φkIt (t) is kth
The instantaneous phase of a component;R (t) is noise or error, and K indicates the component number of signal.The synchronous wavelet transformation that squeezes passes through carefully
The time-frequency curve for changing wavelet transformation, effectively extracts the amplitude factor A of each componentk(t) and instantaneous frequency φ 'k(t) (k=1,
2,…,K).As a kind of special recombination method, the synchronous wavelet transformation SWT that squeezes is according to the coefficient W after wavelet transformationf(a,b)
In the local property of (a, b) nearby by coefficient Wf(a, b) is reassigned to the difference (ω in time frequency planef(a,b),b)(ωf
(a, b) indicates instantaneous frequency of the signal at (a, b)), to keep time-frequency curve thinner apparent, improves frequency resolution and subtract
Small modal overlap, so that precision when component signal reconstructs is higher.
The synchronous wavelet transformation that squeezes gives wavelet mother function ψ (t), signal f based on the continuous wavelet transform of signal
(t) continuous wavelet transform is
In formula: ψ * indicates the conjugation of mother wavelet function, and a is scale factor, and b is shift factor.It is fixed according to Plancherel
Reason, formula (2) are in the equivalence transformation of frequency domain
In formula, ξ is pi,It is the Fourier transformation of f (t), ψ (t) respectively.To simplest signal f
(t)=Acos (ω t), Fourier transformation areAccording to formula (3), continuous wavelet becomes
It is changed to
It is assumed that wavelet function ψ has rapid decay, andIn ξ=ω0Locate integrated distribution, then the wavelet coefficient of signal
Wf(a, b) will be in scalePlace concentrates, but can be in a certain range along size distribution.Therefore, existing research shows
In time-frequency figure, wavelet coefficient Spectral structure range is wider and obscurity boundary, for more complex multicomponent data processing, component signal
Often there is serious spectral aliasing between wavelet coefficient spectrogram.
For any one in wavelet transform result " time-scale " point (b, a).It can be estimated by wavelet coefficient derivation
The instantaneous frequency of signal, i.e.,
Document studies have shown that although wavelet coefficient Wf(a, b) is distributed on each scale a, but no matter what value a takes,
Oscillating characteristic of the wavelet coefficient on shift factor b is directed to instantaneous frequency ωf(a,b).Therefore, the synchronous wavelet transformation that squeezes exists
On the basis of instantaneous frequency, (a, b) → [ω is establishedf(a, b), b] mapping, by wavelet coefficient Wf(a, b) by " time-scale "
Plane is transformed into " time-frequency " plane Wf[ωf(a,b),b].It is squeezed in wavelet transformation SWT synchronous, by any center frequency
Rate ωlNeighbouring sectionWavelet coefficient values be expressed to centre frequency ωlOn, it obtains synchronous squeeze and becomes
Change value Tf(ωl, b), reach raising frequency resolution, reduces the purpose of spectral aliasing.It is practical to calculate, due to a, b, ω from
It dissipates, it is assumed that ai-ai-1=(Δ a)i, synchronous to squeeze wavelet transformation value Tf(ωl, b) and it is represented by
Wavelet transformation SWT and wavelet transformation WT, in short-term Fourier's change are squeezed by the way that a time varying signal f (t) is relatively more synchronous
Change the difference of STFT, f (t) is formed by the Signal averaging of three different frequencies: 0-0.7s is the cosine signal f of 20Hz1(t)=
Cos (40 π t), 0.3-1s are the cosine signal f of 30Hz2(t)=cos (60 π t), 0-1s are the cosine that frequency is vibrated in 80Hz
FM signal f3(t)=cos [160 π t-5cos (30t)], while the white Gaussian noise that signal-to-noise ratio is 3dB is added to signal.It is right
Signal after noise is added is utilized respectively empirical mode decomposition (empirical mode decomposition, EMD), wavelet packet
It converts (wavelet packet transform, WPT) and carries out spectrum analysis with synchronous extruding wavelet transformation SWT, analyze result
As shown in Figure 1.As can be seen that the time-frequency figure of SWT transformation more focuses, is apparent, compared with EMD and WPT, there is higher frequency
Resolution ratio and temporal resolution, and the time-frequency spectrum of each component signal can be accurately extracted using the synchronous wavelet transformation that squeezes.Such as
It in Fig. 1 plus makes an uproar shown in the time frequency analysis of cosine frequency modulation signals, wherein Fig. 1 (a) is the time frequency analysis of EMD, and Fig. 1 (b) is wavelet packet
Time frequency analysis, Fig. 1 (c) is the time frequency analysis of SWT.
The synchronous wavelet transformation that squeezes is reversible, and for multicomponent data processing, passes through Tf(ωl, b) can not only reconstruct it is original
Signal f (t), and can be with each component signal of Accurate Reconstruction fk(t).Assuming that LkIt (t) is in time-frequency figure with fk(t) crestal line
Centered on a minizone, then fk(t) reconstruction formula is
In formulaEMD, wavelet package transforms and SWT are compared it is found that as shown in Figure 1, EMD and small echo
Packet decomposes the accurate division that can not achieve frequency band, and the intersubband decomposed often has serious spectral aliasing and energy is let out
Leakage.UHF PD signal is decomposed according to EMD or wavelet package transforms, obtained subband signal can not be truely and accurately anti-
Reflect the time-domain information in UHF PD signal local segments, the Analysis On Multi-scale Features parameter extracted from each subband signal will necessarily be by
The influence of intersubband spectral aliasing and energy leakage causes characteristic quantity that cannot accurately describe the information that UHF PD signal is included,
It is unfavorable for the identification of subsequent defective type.And accurate careful division, the subband decomposed can be carried out to time frequency plane using SWT
Between be not present spectral aliasing and energy leakage, the multi-scale parameters extracted from each subband signal can accurately describe UHF PD letter
Number time-frequency characteristics.Therefore, UHF PD signal is handled using SWT herein.
Step 2: the multiple dimensioned arrangement entropy measure of SWT
Since the physical essence that different types of insulation defect generates PD is different, different types of discharge pulse can be generated,
So that necessarily there is also larger differences for the time domain waveform for exciting the UHF electromagnetic wave of generation and frequency domain energy distribution;And same kind
Insulation defect electric discharge physical process and discharge pulse excitation uhf electromagnetic wave have stronger similitude.Due to single
The time domain or frequency domain character parameter of scale are vulnerable to external interference, therefore, using the multiple dimensioned energy-distributing feature of time-frequency domain as area
Divide the characteristic quantity of different type defect.And the part spy of signal difference vibration frequency can be extracted by synchronizing extruding wavelet transformation SWT
Property, therefore SWT arrangement moisture in the soil value is calculated just it can be found that exception small and very brief in signal.In the multiple dimensioned table of local discharge signal
On the basis of showing, it is defined as follows and squeezes small echo arrangement entropy measure along the synchronous of size distribution.
If synchronized extruding window Fourier transform (the synchrosqueezing window of local discharge signal
Fourier Transform, SWFT) after in accumulate mode class function be IMTk(k=1,2 ..., K), to mode class function IMTk
Phase space reconfiguration is carried out, can be obtained
In formula: m, τ respectively indicate Embedded dimensions and delay time;The IMT that Q=N- (m-1) τ, N are indicatedkLength.It should
Q reconstruct component then can be obtained as a reconstruct component in every a line of matrix.To in (1) formula j-th of component [IMT (j),
IMT] (j+ τ) ..., IMT (j+ (m-1) τ)], its element is rearranged by increasing mode, if i1,i2,…,imExpression is arranged again
The index of each element position after column has IMT [j+ (i1-1)τ]≤IMT[j+(i2-1)τ]≤…≤IMT[j+(im-1)
τ],
If the value there are two component is equal, i.e.,
IMT[j+(i1- 1) τ]=IMT [j+ (i2-1)τ]
Then in arrangement according to index value i1And i2Size arrange, that is, work as i1< i2When, IMT [j+ (i1- 1) τ] it comes
IMT[j+(i2- 1) τ] before, put in order at this time for
IMT[j+(i1-1)τ]≤IMT[j+(i2-1)τ]
Therefore, for IMTkObtained restructuring matrix after rearranging to each of these row, can obtain
One group based on the symbol sebolic addressing to put in order
S (r)=(i1,i2,…,im), wherein r=1,2 ..., q, and q≤m!
When each row element in restructuring matrix is arranged, if not considering the size arbitrary arrangement of element value,
Its m element shares m!Middle aligning method, i.e., shared m!A symbol sebolic addressing (i1,i2,…,im), it is arranged by element value size
Symbol sebolic addressing S (r) afterwards is one of which.Summarize time that each symbol sebolic addressing S (r) occurs in restructuring matrix arrangement
Number, and calculate its corresponding probability, it is assumed that it is respectively P1,P2,…,Pq, then can according to the definition of Shannon entropy calculate in accumulate
Mode class function IMTkQ kind distinct symbols sequence arrangement entropy Ep(m), i.e.,
When the arrangement S (r) of restructuring matrix disperse the most namely eachWhen, arrangement entropy reaches maximum value Ep(m)
=ln (m!).Therefore, entropy E is arrangedp(m) size accumulates mode class function IMT in can describingkThe random degree of middle sequence: arrangement
Entropy Ep(m) value is smaller, shows IMTkIn data it is more regular;Arrange entropy Ep(m) value is bigger, then shows IMTkIn data
More irregularly, closer to random sequence.Arrange entropy Ep(m) variation, which can reflect and be exaggerated in multiple dimensioned, accumulates mode class letter
Number IMTkThe variations in detail of middle data sequence can go out the mutation of sequence with accurate detection, can get in shelf depreciation classification
Higher resolution ratio.For the ease of using, it usually needs to Ep(m) it is normalized, evenThen 0≤Ep
≤1。
Step 3: carrying out local discharge signal Classifcation of flaws using support vector machines according to the arrangement entropy of extraction
The identification of local discharge signal is real using support vector machines (Support Vector Machine, SVM) classifier
Existing, SVM is a kind of new machine learning method to grow up on the basis of Statistical Learning Theory, it avoids artificial neural network
It the problems such as network structure selection, overfitting and deficient study of network method and local minimum, is taken between study precision and generalization ability
Good balance was obtained, the problems such as suitable for solving pattern classification and regression analysis under higher-dimension, small sample, nonlinear situation.
The SVM kernel function used in this patent is gaussian radial basis function, is handed over using one-to-one more disaggregated models, and using 2- folding
Fork verification method determines the best rule coefficient and kernel functional parameter of SVM.To the data test of 4 kinds of shelf depreciation types, choosing
A part of data are taken to be used to train, another part is used to test.The intersubband decomposed due to SWT there is no spectral aliasing and
The Analysis On Multi-scale Features amount of energy leakage, acquisition can describe more accurately the time-frequency characteristics of original signal and SVM is avoided
The problems such as network structure selection, overfitting and deficient study of Artificial Neural Network and local minimum, thus shelf depreciation is believed
Number Classifcation of flaws achieves better recognition effect.
The following are a specific examples, are known based on the synchronous shelf depreciation type for squeezing wavelet transformation and Analysis On Multi-scale Features parameter
It is other that specific step is as follows:
(1) discharge signal acquires
According to the insulation system feature of inside transformer, shelf depreciation be mainly divided into suspended discharge P1, needle plate electric discharge P2,
Tetra- seed type of creeping discharge P3 and bubble-discharge P4, all types of discharging models are as shown in Figure 2.All disc electrode diameters are
80mm, with a thickness of 10mm, all callipers are 1mm.Wherein Fig. 2 (a) is the electrode structure of suspended discharge in simulation oil, epoxy
Edges of boards edge places the metallic particles that a diameter is 0.3mm;Fig. 2 (b) is the needle plate pole structure of corona discharge in simulation oil, and needle neck is straight
Diameter is 0.2mm, and the epoxy plate between needle and plate electrode is with a thickness of 0.5mm, diameter 1mm;Creeping discharge in Fig. 2 (c) simulation oil;Figure
The model structure of 2 (d) analog insulation internal air gaps electric discharge, air gap is by three layers of diameter for 60mm, with a thickness of the epoxy board group of 1mm
At the Circularhole diameter at center is 20mm.Four kinds of discharging models are both placed in the fuel tank equipped with transformer oil real shown in table 1
It tests under the conditions of room and voltage is applied to every kind of discharging model.Suspended discharge test voltage is 15kV and 24kV in table 1, is expressed as 15/
24, corresponding test sample number is expressed as 15/15.Use detection frequency bandwidth be the high frequency sensors of 0.5~16MHz with
TWPD-ZE Analysis of Partial Discharge instrument carries out discharge signal sampling, and highest sample frequency is 20MHz, sampling time 20ms.
The test condition of Tab.1 partial discharge model
Electric discharge type | Test voltage/kV | Number of samples |
Bubble-discharge | 10/15 | 40/40 |
Creeping discharge | 15/20 | 40/40 |
Needle plate electric discharge | 10/15 | 40/40 |
Suspended discharge | 15/24 | 40/40 |
Apply voltage to every kind of partial discharge model in laboratory conditions to show using pulse current method using high-performance
Wave device acquires local discharge signal.In order to avoid the randomness of test, every kind of electric discharge type has made 40 samples.
(2) feature extraction
It selects Morlet small echo as wavelet basis, extruding point is synchronized to the UHF PD signal of collected 4 kinds of defects
Solution, is divided into 10 subbands for SWT time-frequency spectrum, utilizes SWT inverse transformation expression re-formation component signal to each subband.From 4 kinds of UHF PD
50 groups of data are randomly selected in sample of signal library respectively as training sample, the remaining 30 groups of data of every kind of electric discharge type are as survey
Sample sheet completes the feature extraction to all UHF PD signals according to step shown in Fig. 3, obtains 4 kinds of UHF PD signal characteristics
Library.Statistical analysis obtains 95% confidence interval of the multiple dimensioned arrangement spectrum entropy feature of 4 kinds of UHF PD signals, as shown in Figure 4.
As shown in Figure 4, the multiple dimensioned arrangement entropy of bubble-discharge and the UHF PD signal of creeping discharge is concentrated mainly on
The frequency range of 400MHz or more, and the arrangement entropy of the UHF PD signal of needle plate electric discharge and suspended discharge is concentrated mainly on 400MHz or less
Frequency range.Due to the waveform and steepness difference of different types of discharge pulse, so as to cause the UHF signal arrangement entropy cloth tool of excitation
There is biggish difference, illustrates that it is feasible for carrying out Classifcation of flaws using multiple dimensioned arrangement entropy feature.
Therefore, there are notable differences for the subband arrangement entropy feature of 4 kinds of discharging model local discharge superhigh frequencies, can use office
Portion's electric discharge ultra-high frequency signal characteristic parameter carries out PD Pattern Recognition.
(3) Classifcation of flaws
The identification of UHF PD signal using support vector machines (Support Vector Machine, SVM) classifier realize,
SVM is a kind of new machine learning method to grow up on the basis of Statistical Learning Theory, it avoids artificial neural network
It the problems such as network structure selection, overfitting and deficient study of method and local minimum, is obtained between study precision and generalization ability
Good balance, the problems such as suitable for solving pattern classification and regression analysis under higher-dimension, small sample, nonlinear situation.It is real
The most common gaussian radial basis function of Selection of kernel function of middle SVM is tested, expression formula is
In identification process, SVM uses one-to-one more disaggregated models, and determines SVM most using 2- folding cross validation method
Good regularization coefficient C=0.3 and kernel functional parameter σ=0.65.To 80 groups of experimental datas of every kind of electric discharge type, 50 groups of use are chosen
It trains, 30 groups are used to test.Recognition result is as shown in table 2, and unit is % in table.
2 shelf depreciation recognition result of table
Be respectively adopted EMD and WPT UHF PD signal is decomposed (using db10 small echo as wavelet basis when WAVELET PACKET DECOMPOSITION,
7) Decomposition order is set as, extract identical characteristic quantity, comparison recognition result discovery, using the decomposition side SWT to every layer of decomposition coefficient
The discrimination that method obtains is apparently higher than EMD method and WAVELET PACKET DECOMPOSITION method.This is because the intersubband that SWT is decomposed is not deposited
In spectral aliasing and energy leakage, the Analysis On Multi-scale Features amount of acquisition can describe the time-frequency characteristics of original signal more accurately,
Thus achieve better recognition effect.Multiple dimensioned energy feature and multiple dimensioned Energy Spectrum Entropy feature all achieve preferable identification knot
Fruit, average recognition rate are higher than 90%.
Method of the invention overcomes wavelet decomposition intersubband by the synchronous decomposition for squeezing transformation to local discharge signal
The shortcomings that there are spectral aliasing and energy leakages, the multiple dimensioned arrangement entropy characteristic parameter used can effectively portray UHFPD signal
In the Energy distribution and complexity information of time-frequency domain, there is preferable stability and anti-interference ability.Method of the invention is applicable in
Transformer partial discharge signal identification in all kinds of situations.The local discharge signal of method of the invention is averaged recognition accuracy
Higher than 90%, hence it is evident that be better than other common partial discharge of transformer recognition methods.
Effect:
Synchronous squeeze wavelet transformation (Synchrosqueezing wavelet transform, SWT) is in wavelet transformation
On the basis of the new Time-Frequency Analysis Method of one kind that grows up, it is based on continuous wavelet transform, by wavelet coefficient
It is recombinated, therefrom extracts time-frequency curve, therefore there is high precision and frequency resolution.Intersubband after SWT is decomposed does not have
There is cross term, spectral aliasing and energy leakage is not present in intersubband, and subband signal can accurately describe the time-frequency characteristics of original signal,
It has obtained widely answering in fields such as time varying signal spectrum analysis, Earthquake signal detection, Sonar Signal analysis and mechanical fault diagnosis
With.4 kinds of typical transformer partial discharge signals are handled using SWT, from UHF PD signal in the multiple dimensioned of time-frequency domain
The difference of arrangement entropy distribution is set out, and research can effectively distinguish the Analysis On Multi-scale Features parameter of different insulative defect, and using support
Vector machine classifier realizes electric discharge type identification.
But existing research is to the identification based on the synchronous shelf depreciation for squeezing wavelet transformation and Analysis On Multi-scale Features parameter
It is not too many in terms of patent to be related to.
In view of the above problems, this patent is proposed based on the synchronous shelf depreciation for squeezing wavelet transformation and Analysis On Multi-scale Features parameter
The identification of classification squeezes the high-precision Time-Frequency Analysis function of wavelet transformation and to the robustness of noise, proposition one using synchronous
Kind is based on the synchronous local discharge characteristic extracting method for squeezing wavelet transformation.Firstly, using the synchronous wavelet transformation that squeezes to 4 kinds of allusion quotations
The transformer partial discharge signal of type is decomposed, and to overcome real WAVELET PACKET DECOMPOSITION intersubband, there are spectral aliasings and energy leakage
Defect;Then, using the difference of local discharge signal energy and complexity on different decomposition scale, multiple dimensioned arrangement is utilized
The characteristic quantity that entropy is identified as electric discharge type;Finally, the characteristic quantity support vector machine classifier extracted is carried out discharge mode
Identification.The experimental results showed that this method can obtain recognition effect more better than EMD and WAVELET PACKET DECOMPOSITION, it was demonstrated that this method
Validity.
Claims (2)
1. a kind of based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation characterized by comprising
Step 1: decomposing using the synchronous wavelet transformation that squeezes to 4 kinds of typical transformer partial discharge signals, synchronization is utilized
It squeezing wavelet transformation and accurate careful division is carried out to time frequency plane, decomposition obtains subband signal,
Generally can be analyzed to the superposition of multiple eigenfunctions, i.e. signal f (t) is represented by time varying signal f (t)
In formula: AkIt (t) is the instantaneous amplitude of k-th of component;AkIt (t) is the instantaneous amplitude of k-th of component, φkIt (t) is k-th point
The instantaneous phase of amount;R (t) is noise or error, and K indicates the component number of signal, and synchronous extruding wavelet transformation is small by refining
The time-frequency curve of wave conversion extracts the amplitude factor A of each componentk(t) and instantaneous frequency φ 'k(t) (k=1,2 ..., K), together
Step squeezes wavelet transformation according to the coefficient W after wavelet transformationf(a, b) is in (a, b) neighbouring local property by coefficient Wf(a, b) weight
Difference (the ω being newly assigned in time frequency planef(a,b),b)(ωf(a, b) indicates instantaneous frequency of the signal at (a, b)), thus
Keep time-frequency curve thinner apparent, improve frequency resolution and reduce modal overlap, so that precision when component signal reconstructs is more
Height,
The synchronous wavelet transformation that squeezes gives wavelet mother function ψ (t) based on the continuous wavelet transform of signal, signal f's (t)
Continuous wavelet transform is
In formula: ψ * indicates the conjugation of mother wavelet function, and a is scale factor, and b is shift factor, according to Plancherel theorem, formula
(2) it is in the equivalence transformation of frequency domain
In formula, ξ is pi,The Fourier transformation of f (t), ψ (t) respectively, to simplest signal f (t)=
Acos (ω t), Fourier transformation areAccording to formula (3), continuous wavelet transform is
It is assumed that wavelet function ψ has rapid decay, andIn ξ=ω0Locate integrated distribution, then the wavelet coefficient W of signalf(a,
It b) will be in scalePlace concentrates, but can in a certain range along size distribution,
For any one in wavelet transform result " time-scale " point, (b a) can estimate signal by wavelet coefficient derivation
Instantaneous frequency, i.e.,
The synchronous wavelet transformation that squeezes establishes (a, b) → [ω on the basis of instantaneous frequencyf(a, b), b] mapping, by wavelet systems
Number Wf(a, b) is transformed into " time-frequency " plane W by " time-scale " planef[ωf(a, b), b], by any centre frequency
ωlNeighbouring sectionWavelet coefficient values be expressed to centre frequency ωlOn, it obtains synchronous squeeze and converts
Value Tf(ωl, b), in calculating, due to a, b, ω is discrete, it is assumed that ai-ai-1=(Δ a)i, synchronous to squeeze wavelet transformation value Tf
(ωl, b) and it is represented by
Wavelet transformation and wavelet transformation WT, Short Time Fourier Transform STFT are squeezed by the way that time varying signal f (t) is relatively more synchronous
Difference, f (t) are formed by the Signal averaging of three different frequencies: 0-0.7s is the cosine signal f of 20Hz1(t)=cos (40 π
T), 0.3-1s is the cosine signal f of 30Hz2(t)=cos (60 π t), 0-1s are the cosine frequency modulation letter that frequency is vibrated in 80Hz
Number f3(t)=cos [160 π t-5cos (30t)], while the white Gaussian noise that signal-to-noise ratio is 3dB is added to signal, it makes an uproar to addition
Signal after sound is utilized respectively empirical mode decomposition, and wavelet package transforms carry out spectrum analysis with synchronous extruding wavelet transformation,
The synchronous wavelet transformation that squeezes is reversible, and for multicomponent data processing, passes through Tf(ωl, b) and it can not only reconstruct original signal f
It (t), and can be with each component signal of Accurate Reconstruction fk(t), it is assumed that LkIt (t) is in time-frequency figure with fk(t) centered on crestal line
A minizone, then fk(t) reconstruction formula is
In formula
Step 2: synchronous squeeze the multiple dimensioned arrangement entropy measure of wavelet transformation
It is synchronous to squeeze wavelet transformation using the multiple dimensioned energy-distributing feature of time-frequency domain as the characteristic quantity for distinguishing different type defect
The local characteristics of signal difference vibration frequency are extracted, calculate the synchronous wavelet transformation arrangement moisture in the soil value that squeezes just it can be found that micro- in signal
Small and very brief exception is defined as follows and squeezes along the synchronous of size distribution on the basis of local discharge signal multi-scale Representation
Small echo arranges entropy measure,
If accumulateing mode class function in after the synchronized extruding window Fourier transform of local discharge signal is IMTk(k=1,2 ...,
K), to mode class function IMTkPhase space reconfiguration is carried out, can be obtained
In formula: m, τ respectively indicate Embedded dimensions and delay time;The IMT that Q=N- (m-1) τ, N are indicatedkLength, by the matrix
Q reconstruct component then can be obtained as a reconstruct component in every a line, to j-th of component [IMT (j), IMT] (j in (1) formula
+ τ) ..., IMT (j+ (m-1) τ)], its element is rearranged by increasing mode, if i1,i2,…,imExpression is each after rearranging
The index of element position has IMT [j+ (i1-1)τ]≤IMT[j+(i2-1)τ]≤…≤IMT[j+(im- 1) τ],
If the value there are two component is equal, i.e.,
IMT[j+(i1- 1) τ]=IMT [j+ (i2-1)τ]
Then in arrangement according to index value i1And i2Size arrange, that is, work as i1< i2When, IMT [j+ (i1- 1) τ] come IMT [j
+(i2- 1) τ] before, put in order at this time for
IMT[j+(i1-1)τ]≤IMT[j+(i2-1)τ]
Therefore, for IMTkObtained restructuring matrix after rearranging to each of these row, can obtain one group
Based on the symbol sebolic addressing to put in order
S (r)=(i1,i2,…,im), wherein r=1,2 ..., q, and q≤m!Each row element in restructuring matrix is arranged
When column, if not considering the size arbitrary arrangement of element value, m element shares m!Middle aligning method, i.e., shared m!A symbol
Number sequence (i1,i2,…,im), the symbol sebolic addressing S (r) after being arranged by element value size is one of which, is summarized each
The number that kind symbol sebolic addressing S (r) occurs in restructuring matrix arrangement, and calculate its corresponding probability, it is assumed that it is respectively P1,
P2,…,Pq, then can according to the definition of Shannon entropy calculate in accumulate mode class function IMTkQ kind distinct symbols sequence row
Column entropy Ep(m), i.e.,
When the arrangement S (r) of restructuring matrix disperse the most namely eachWhen, arrangement entropy reaches maximum value Ep(m)=ln
(m!), therefore, arrange entropy Ep(m) size accumulates mode class function IMT in can describingkThe random degree of middle sequence: arrangement entropy Ep
(m) value is smaller, shows IMTkIn data it is more regular;Arrange entropy Ep(m) value is bigger, then shows IMTkIn data more not
Rule arranges entropy E closer to random sequencep(m) variation, which can reflect and be exaggerated in multiple dimensioned, accumulates mode class function
IMTkThe variations in detail of middle data sequence can go out the mutation of sequence with accurate detection, shelf depreciation classification when can get compared with
High resolution ratio, for the ease of using, it usually needs to Ep(m) it is normalized, evenThen 0≤Ep≤
1,
Step 3: carrying out local discharge signal Classifcation of flaws using support vector machines according to the arrangement entropy of extraction.
2. as described in claim 1 based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation, it is characterised in that:
In step 3, the most common gaussian radial basis function of the Selection of kernel function of SVM, expression formula is
In identification process, SVM uses one-to-one more disaggregated models, and the best rule of SVM are determined using 2- folding cross validation method
Then change coefficient C=0.3 and kernel functional parameter σ=0.65.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810829627.6A CN108983051B (en) | 2018-07-25 | 2018-07-25 | Partial discharge type identification method based on synchronous extrusion wavelet transform |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810829627.6A CN108983051B (en) | 2018-07-25 | 2018-07-25 | Partial discharge type identification method based on synchronous extrusion wavelet transform |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108983051A true CN108983051A (en) | 2018-12-11 |
CN108983051B CN108983051B (en) | 2021-02-09 |
Family
ID=64551338
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810829627.6A Expired - Fee Related CN108983051B (en) | 2018-07-25 | 2018-07-25 | Partial discharge type identification method based on synchronous extrusion wavelet transform |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108983051B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109668733A (en) * | 2018-12-21 | 2019-04-23 | 苏州大学 | Variation nonlinear model decomposes variable speed Method for Bearing Fault Diagnosis |
CN109709448A (en) * | 2019-03-06 | 2019-05-03 | 南京工程学院 | A kind of power distribution network single-phase high-impedance selection method based on synchronous extruding wavelet transformation |
CN110133538A (en) * | 2019-05-16 | 2019-08-16 | 合肥工业大学 | A kind of ANPC three-level inverter open-circuit fault diagnostic method and experiment porch |
CN110161123A (en) * | 2019-06-21 | 2019-08-23 | 南昌航空大学 | A kind of new defect inspection method based on magnetic striction wave guide |
CN110244202A (en) * | 2019-06-21 | 2019-09-17 | 三峡大学 | Based on synchronous compression wavelet transformed domain partial discharge of transformer denoising method |
CN110464517A (en) * | 2019-08-16 | 2019-11-19 | 杭州电子科技大学 | Electromyography signal recognition methods based on Wavelet-Weighted arrangement entropy |
EP3699614A1 (en) * | 2019-02-22 | 2020-08-26 | Ormazabal Corporate Technology, A.I.E. | Method and system of partial discharge recognition for diagnosing electrical networks |
CN111735533A (en) * | 2020-06-08 | 2020-10-02 | 贵州电网有限责任公司 | Transformer direct-current magnetic bias judgment method based on vibration signal wavelet energy spectrum characteristics |
CN111781439A (en) * | 2020-05-28 | 2020-10-16 | 广西电网有限责任公司梧州供电局 | Power cable partial discharge signal detection method and device |
CN112348782A (en) * | 2020-10-27 | 2021-02-09 | 天津大学 | Change detection method based on plurality of controllable pyramids |
CN112686182A (en) * | 2021-01-04 | 2021-04-20 | 华北电力大学(保定) | Partial discharge mode identification method and terminal equipment |
CN112698158A (en) * | 2020-11-13 | 2021-04-23 | 西安交通大学 | Partial discharge electrical positioning method and device for transformer |
CN114062995A (en) * | 2021-11-15 | 2022-02-18 | 通号(长沙)轨道交通控制技术有限公司 | Mutual inductor fault diagnosis method, equipment and medium based on electric quantity multi-feature fusion |
CN114113943A (en) * | 2021-11-25 | 2022-03-01 | 广东电网有限责任公司广州供电局 | Transformer partial discharge detection system, method and equipment based on current and ultrasonic signals |
CN114271836A (en) * | 2022-01-25 | 2022-04-05 | 合肥学院 | Intelligent myoelectricity detection processing method and device based on wavelet transformation |
CN115310499A (en) * | 2022-10-12 | 2022-11-08 | 极晨智道信息技术(北京)有限公司 | Industrial equipment fault diagnosis system and method based on data fusion |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140133483A1 (en) * | 2012-11-14 | 2014-05-15 | Broadcom Corporation | Distributed Switch Architecture Using Permutation Switching |
CN105956526A (en) * | 2016-04-22 | 2016-09-21 | 山东科技大学 | Method for identifying a microearthquake event with low signal-to-noise ratio based on multi-scale permutation entropy |
CN107356843A (en) * | 2017-04-17 | 2017-11-17 | 武汉科技大学 | The partial discharge of transformer method for diagnosing faults of small echo is synchronously extruded based on gradient threshold |
CN108038580A (en) * | 2017-12-30 | 2018-05-15 | 国网江苏省电力公司无锡供电公司 | The multi-model integrated Forecasting Methodology of photovoltaic power based on synchronous extruding wavelet transformation |
-
2018
- 2018-07-25 CN CN201810829627.6A patent/CN108983051B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140133483A1 (en) * | 2012-11-14 | 2014-05-15 | Broadcom Corporation | Distributed Switch Architecture Using Permutation Switching |
CN105956526A (en) * | 2016-04-22 | 2016-09-21 | 山东科技大学 | Method for identifying a microearthquake event with low signal-to-noise ratio based on multi-scale permutation entropy |
CN107356843A (en) * | 2017-04-17 | 2017-11-17 | 武汉科技大学 | The partial discharge of transformer method for diagnosing faults of small echo is synchronously extruded based on gradient threshold |
CN108038580A (en) * | 2017-12-30 | 2018-05-15 | 国网江苏省电力公司无锡供电公司 | The multi-model integrated Forecasting Methodology of photovoltaic power based on synchronous extruding wavelet transformation |
Non-Patent Citations (1)
Title |
---|
喻敏等: "同步挤压小波变换在电力系统低频振荡模态参数提取中的应用", 《电工技术学报》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109668733A (en) * | 2018-12-21 | 2019-04-23 | 苏州大学 | Variation nonlinear model decomposes variable speed Method for Bearing Fault Diagnosis |
EP3699614A1 (en) * | 2019-02-22 | 2020-08-26 | Ormazabal Corporate Technology, A.I.E. | Method and system of partial discharge recognition for diagnosing electrical networks |
CN109709448A (en) * | 2019-03-06 | 2019-05-03 | 南京工程学院 | A kind of power distribution network single-phase high-impedance selection method based on synchronous extruding wavelet transformation |
CN110133538A (en) * | 2019-05-16 | 2019-08-16 | 合肥工业大学 | A kind of ANPC three-level inverter open-circuit fault diagnostic method and experiment porch |
CN110244202B (en) * | 2019-06-21 | 2021-07-20 | 三峡大学 | Transformer partial discharge denoising method based on synchronous compression wavelet transform domain |
CN110244202A (en) * | 2019-06-21 | 2019-09-17 | 三峡大学 | Based on synchronous compression wavelet transformed domain partial discharge of transformer denoising method |
CN110161123A (en) * | 2019-06-21 | 2019-08-23 | 南昌航空大学 | A kind of new defect inspection method based on magnetic striction wave guide |
CN110464517A (en) * | 2019-08-16 | 2019-11-19 | 杭州电子科技大学 | Electromyography signal recognition methods based on Wavelet-Weighted arrangement entropy |
CN110464517B (en) * | 2019-08-16 | 2021-09-07 | 杭州电子科技大学 | Electromyographic signal identification method based on wavelet weighted arrangement entropy |
CN111781439A (en) * | 2020-05-28 | 2020-10-16 | 广西电网有限责任公司梧州供电局 | Power cable partial discharge signal detection method and device |
CN111735533A (en) * | 2020-06-08 | 2020-10-02 | 贵州电网有限责任公司 | Transformer direct-current magnetic bias judgment method based on vibration signal wavelet energy spectrum characteristics |
CN111735533B (en) * | 2020-06-08 | 2022-05-13 | 贵州电网有限责任公司 | Transformer direct-current magnetic bias judgment method based on vibration signal wavelet energy spectrum characteristics |
CN112348782A (en) * | 2020-10-27 | 2021-02-09 | 天津大学 | Change detection method based on plurality of controllable pyramids |
CN112348782B (en) * | 2020-10-27 | 2022-03-29 | 天津大学 | Change detection method based on plurality of controllable pyramids |
CN112698158A (en) * | 2020-11-13 | 2021-04-23 | 西安交通大学 | Partial discharge electrical positioning method and device for transformer |
CN112698158B (en) * | 2020-11-13 | 2022-04-12 | 西安交通大学 | Partial discharge electrical positioning method and device for transformer |
CN112686182A (en) * | 2021-01-04 | 2021-04-20 | 华北电力大学(保定) | Partial discharge mode identification method and terminal equipment |
CN112686182B (en) * | 2021-01-04 | 2023-12-26 | 华北电力大学(保定) | Partial discharge mode identification method and terminal equipment |
CN114062995A (en) * | 2021-11-15 | 2022-02-18 | 通号(长沙)轨道交通控制技术有限公司 | Mutual inductor fault diagnosis method, equipment and medium based on electric quantity multi-feature fusion |
CN114113943A (en) * | 2021-11-25 | 2022-03-01 | 广东电网有限责任公司广州供电局 | Transformer partial discharge detection system, method and equipment based on current and ultrasonic signals |
CN114271836A (en) * | 2022-01-25 | 2022-04-05 | 合肥学院 | Intelligent myoelectricity detection processing method and device based on wavelet transformation |
CN114271836B (en) * | 2022-01-25 | 2023-08-29 | 合肥学院 | Intelligent myoelectricity detection processing method and device based on wavelet transformation |
CN115310499A (en) * | 2022-10-12 | 2022-11-08 | 极晨智道信息技术(北京)有限公司 | Industrial equipment fault diagnosis system and method based on data fusion |
Also Published As
Publication number | Publication date |
---|---|
CN108983051B (en) | 2021-02-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108983051A (en) | Based on the synchronous shelf depreciation kind identification method for squeezing wavelet transformation | |
Ma et al. | Interpretation of wavelet analysis and its application in partial discharge detection | |
Lee et al. | New fault diagnosis of circuit breakers | |
CN108426713A (en) | Rolling bearing Weak fault diagnostic method based on wavelet transformation and deep learning | |
Liu et al. | Deep learning and recognition of radar jamming based on CNN | |
Contin et al. | Classification and separation of partial discharge signals by means of their auto-correlation function evaluation | |
CN106405339A (en) | Power transmission line fault reason identification method based on high and low frequency wavelet feature association | |
CN108229382A (en) | Vibration signal characteristics extracting method, device, storage medium and computer equipment | |
CN110146812A (en) | A kind of Method of Motor Fault Diagnosis based on the study of characteristic node increment type width | |
Allahbakhshi et al. | A method for discriminating original pulses in online partial discharge measurement | |
CN103983850A (en) | Power system harmonious wave compressed signal reconstruction and detection method based on compressed sensing | |
CN107561416A (en) | A kind of local discharge signal acquisition system and method based on compressed sensing | |
CN106682615A (en) | Method for detecting underwater dim small target | |
CN109085468A (en) | A kind of recognition methods of cable local discharge insulation defect | |
CN106291281A (en) | A kind of substation equipment shelf depreciation alignment system and method thereof | |
CN103335841A (en) | An antifriction bearing fault diagnosis method employing a pulse wavelet energy spectrum analysis | |
CN106546818A (en) | A kind of harmonic signal detection method based on DNL Mode Decomposition | |
Song et al. | A method for specific emitter identification based on empirical mode decomposition | |
CN107340055A (en) | It is a kind of based on the random resonant weak signal detection method for estimating fusion more | |
Zhongrong et al. | Application of complex wavelet transform to suppress white noise in GIS UHF PD signals | |
Xue et al. | Application of feature extraction method based on 2D—LPEWT in cable partial discharge analysis | |
CN112881879A (en) | High-voltage cable terminal partial discharge mode identification method, device and equipment | |
Uckol et al. | Identification of corona discharges based on wavelet scalogram images with deep convolutional neural networks | |
CN103528672A (en) | Magneto acoustic signature analysis method based on spectral analysis and time domain correlation method | |
Boczar et al. | Analysis of the acoustic emission pulses generated by partial electrical discharges |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210209 Termination date: 20210725 |